[trissim]

Headless dataclass analysis currently discovers presentation documentation while admitting values. Existing ParameterInfo now owns the canonical parameter declaration and derives dataclass descriptions on help demand; SignatureAnalyzer retains its existing declaration/source cache and prepares the registered schema-source closure without evaluating factories or full type analysis. Value/default factory admission and callable type-bearing documentation remain on their existing paths. Description-preserving _replace does not demand help.

Public0.2 migration: ParameterInfo keeps its five named fields, constructor and _replace, but is a frozen nominal declaration rather than NamedTuple; tuple unpacking/_asdict, UnifiedParameterInfo/source_type and analyze_nested are removed. Dataclass presentation failures yield absent help without discarding valid declarations. Existing installed PyQtReactive/MetaclassRegistry consumer audit found no removed-API uses; PyQt's direct parameter-help consumer reads the supported description property.

Validation: existing source/signature/declaration controls, factory/error/cache and demanded help/reset parity passed. Current CI passes Python3.11/3.12 on Linux/Windows/macOS. Three Python3.10 recursive AbstractSequence decoder tests fail with140passing: dataclass_projection.py is byte-identical to current main430 (SHA256d2be277051b070d73503025a3addcde8016787e18a369c45b60be0c0edb65fea), so this is a documented preexisting decoder defect, not a change in this PR. No decoder patch or qualified-wheel replacement is included. Separate follow-up: issue13.

Current production qualification (2026-10-05): OpenHCS integration92c22 used the exact private PythonIntrospect0.2.0/ObjectState1.3.0 wheels, normal public RegistryService READY before clocks, and default internal runtime profiling OFF. One Illum3 observation measured compile206.535ms, execution264.932ms, total565.896ms. Both older retained native output sets passed the full nonempty-image/inventory/source-association comparison under the existing1e-6 policy (maximum pixel difference2.98e-8). Source/input/helper/eight-dependency/four-native-ABI guards passed; owned processes terminated.

This is an absolute candidate observation, not a controlled ON→OFF or individual-patch gain. Older unpaired native full-invocation mean695.522ms gives1.229×total; the2×total target remains218.135ms away. Native module mean626.957ms is a different clock. Evidence: /home/ts/.local/state/openhcs-maintenance/20261005/parameter-owner-illum3-default-ordinary-v1/receiving-summary.json; durable OpenHCS receipt: benchmark/results/perf_parameter_declaration_first_job_20261005/ordinary-observer-off.json.

Earlier evidence is retained with its limits: authentic first-job PipelineConfig+10-step saved admission129.320→82.766ms (46.555ms reduction, mainly PipelineConfig); already-warm second-case3D231.263→235.966ms showed no gain. The earlier standalone261ms3D replay is not a whole-compiler gain. Failed helper attempts and the prior negative warmed ordinary observation remain archived; no repeat timing was used to rescue them.

Release state: this PR merged as43cdbf3; its tree is identical to qualified2ab8140. Root pushed annotatedv0.2.0 and is tracking the supported tag-triggered publisher. ObjectState22 has also merged as7db1519 with identical qualified source and annotatedv1.3.0; publication availability and shared-environment integration remain pending. Private wheels were used only for receiving.

Closes #11

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.